Building Energy Consumption Prediction Using CatBoost With Hybrid Random Search and Bayesian Optimization

Residential buildings are major contributors to global energy consumption, with cooling and heating loads representing a substantial portion of this demand. Accurate estimation of these energy loads is critical for the design of energy-efficient buildings. This study proposes an innovative approach to predict the energy consumption of residential buildings, focusing on cooling load and heating load.
The CatBoost model, optimized through a hybrid technique that combines Random Search and Bayesian Optimization, is employed to enhance prediction accuracy and computational efficiency. The performance of the proposed model is compared with several machine learning algorithms, including SVR, GBM,
Random Forest, AdaBoost, and XGBoost, to assess its effectiveness in estimating energy consumption.
In addition, an analysis of feature importance identifies key input parameters, such as overall height, relative compactness, and roof area that influence the forecasting of cooling and heating loads. Experimental results demonstrate that the proposed hybrid-optimized CatBoost model outperforms all other methods that have targeted the same dataset in the literature, achieving an RMSE of 0.045654, MAE of 0.031149, MSE of 0.002084, and R2 of 0.998102 for cooling load prediction, and an RMSE of 0.024707, MAE of 0.018723, MSEof0.000610, and R2 of 0.999451 for heating load prediction. These findings provide practical insights
for engineers and architects to enhance building design and energy efficiency